The AI Voice Experiment: Yielding to Temptation While Keeping Our Soul

The AI Voice Experiment: Yielding to Temptation While Keeping Our Soul

Oscar Wilde once observed that “the only way to get rid of a temptation is to yield to it.” As we stand at the precipice of the AI revolution in content creation, Wilde’s words have never felt more relevant – or more dangerous.

Wilde’s words in this context make the point that AI tools are very very tempting – they lure us with the promise of “easiness” and we are hungry for ways to be lazy.

This is dangerous because prices for AI tools will rise dramatically soon (the company behind Claude has announced a 400% price rise for its new versions) and, just like a drug dealer getting people hooked on free or cheap samples, we might end up trapped, having lost our skills to write independently of AI support.

Stand by for a long article with plenty of cut and pasted text so you can see all the workings behind the curtain.

The Voice Capture Framework

This post is a companion piece to a webinar I ran yesterday on how to train AI to mimic your voice. I am going to use some of the elements from that workshop in this post, using ourselves as an example.

Our journey began with a fundamental framework for capturing voice – these are the things you need to consider:

Tone: [e.g., friendly, formal, casual, professional]

Style: [e.g., conversational, informative, humorous, technical]

Language and Dialect: [e.g., American English, British English, Australian English]

Speech Patterns: [e.g., uses idioms, avoids jargon, includes metaphors]

Audience: [e.g., children, experts, general public]

Level of Detail: [e.g., concise summaries, in-depth explanations, step-by-step instructions]

Additional Preferences: [e.g., asks rhetorical questions, uses first-person perspective, incorporates storytelling]

First Attempt at Voice Capture

I took the Talking About Marketing Podcast as my guinea pig for the workshop. I am one of the hosts and David Olney is the other one. Together, we have a particular chemistry that I wanted to capture for the purpose of writing the show notes that complement each episode.

My initial attempt to capture our podcast’s voice looked like this:

Tone: Warm, authoritative but driven by an eager sense of curiosity to understand and then share insights with others

Style: Informative, sometimes technical, often infused with some dry humour quips

Language and Dialect: Australian English, but never with Australian slang

Speech Patterns: Strives to use analogies to make things understandable, typically one speaker holds the stage and then throws to the other to seek their opinion, avoids jargon, includes metaphors when needed

Audience: Small business owners or leaders

Level of Detail: Start segments with an overview, then use some specific examples and direct quotes to prompt further discussion about how the topic practically applies to the daily life and work of the listener

Additional Preferences: There should be a relaxed but highly engaged feel to this voice.

The AI Analysis

The test then came when we fed a transcript from Season 1, Episode 9 into Claude AI, and asked it if what it read in that transcript sat well with my first attempt of defining our “voice”. Here’s the transcript sample I shared with it:

So, that was the text we fed it. Now, here is what Claude’s verdict was about how accurate (or otherwise) my attempt at defining our voice was:

The Refined Voice Guide

This led to a more nuanced understanding of our voice. This is the modified version that Claude came up with:

Testing the New Guidelines

The real test came with Season 5, Episode 9. This is a new episode that won’t be released until Tuesday, December 10, 2024, at 3am.

As you’ll see, the first AI-generated show notes were technically accurate but missed our essence:

I then provided this feedback to Claude:

I think we need to make some changes to the voice style so the show notes capture more of our levity and quick wittedness. They are thorough but they are reading a bit dry to me. These show notes should distill the source thought or material but particularly highlight some of the insights that Steve and David have brought to the conversation – that’s the interesting bit, that’s what makes us different from just googling the topic. I will need to take your direction on rewording the source style document about the podcast and the writing style notes.

After providing this feedback, we received this vastly improved version:

The Final Evolution

This journey led us to our most refined voice guide yet:

Tone

Style

Language and Dialect

Speech Patterns

Dynamic Elements (New Section)

Audience

Level of Detail

Additional Preferences

The Paradoxical Truth

Here’s the fascinating paradox: By using AI to help capture our voice, we actually became more aware of what makes our voice distinctive. It’s like using a mirror to perfect your posture – the tool helps you see yourself more clearly.

What once took 45 minutes of careful segment-by-segment processing now takes 60 seconds. This efficiency doesn’t make us lazy; it frees us to focus on what matters most – creating valuable content for our audience.

Oscar Wilde also said, “In matters of grave importance, style, not sincerity, is the vital thing.” In our case, we’ve found that by understanding our style more deeply through this AI experiment, we’ve actually enhanced our sincerity. We’ve yielded to the temptation of AI assistance but maintained our soul in the process.

The key isn’t to resist the AI temptation entirely but to yield to it thoughtfully, using it as a tool for enhancement rather than replacement. After all, the best way to maintain control over a temptation might just be to understand it completely.

Get helpful marketing articles and links to our latest podcast episodes